CoxAIPW
CRANCoxAIPWMaRDI QIDQ76860FDOQ76860
Doubly Robust Inference for Cox Marginal Structural Model with Informative Censoring
Ronghui Xu, Dennis Rava, Jiyu Luo
Last update: 20 September 2023
Copyright license: GNU General Public License, version 3.0
Software version identifier: 0.0.1, 0.0.2, 0.0.3
Doubly robust estimation and inference of log hazard ratio under the Cox marginal structural model with informative censoring. An augmented inverse probability weighted estimator that involves 3 working models, one for conditional failure time T, one for conditional censoring time C and one for propensity score. Both models for T and C can depend on both a binary treatment A and additional baseline covariates Z, while the propensity score model only depends on Z. With the help of cross-fitting techniques, achieves the rate-doubly robust property that allows the use of most machine learning or non-parametric methods for all 3 working models, which are not permitted in classic inverse probability weighting or doubly robust estimators. When the proportional hazard assumption is violated, CoxAIPW estimates a causal estimated that is a weighted average of the time-varying log hazard ratio. Reference: Luo, J. (2023). Statistical Robustness - Distributed Linear Regression, Informative Censoring, Causal Inference, and Non-Proportional Hazards [Unpublished doctoral dissertation]. University of California San Diego.; Luo & Xu (2022) <doi:10.48550/arXiv.2206.02296>; Rava (2021) <https://escholarship.org/uc/item/8h1846gs>.
This page was built for software: CoxAIPW
